ofproto-dpif-rid: Don't carry actset_output explicitly in metadata.
[cascardo/ovs.git] / ofproto / ofproto-dpif.c
index 49001e8..3281eaa 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
+ * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -303,7 +303,7 @@ struct ofproto_dpif {
     /* Special OpenFlow rules. */
     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
-    struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
+    struct rule_dpif *drop_frags_rule; /* Used in OFPUTIL_FRAG_DROP mode. */
 
     /* Bridging. */
     struct netflow *netflow;
@@ -341,9 +341,9 @@ struct ofproto_dpif {
     uint64_t change_seq;           /* Connectivity status changes. */
 
     /* Work queues. */
-    struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
-    struct seq *pins_seq;          /* For notifying 'pins' reception. */
-    uint64_t pins_seqno;
+    struct guarded_list ams;      /* Contains "struct ofproto_async_msgs"s. */
+    struct seq *ams_seq;          /* For notifying 'ams' reception. */
+    uint64_t ams_seqno;
 };
 
 /* All existing ofproto_dpif instances, indexed by ->up.name. */
@@ -400,20 +400,19 @@ ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
     ofproto_flow_mod(&ofproto->up, &ofm);
 }
 
-/* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
- * Takes ownership of 'pin' and pin->packet. */
+/* Appends 'am' to the queue of asynchronous messages to be sent to the
+ * controller.  Takes ownership of 'am' and any data it points to. */
 void
-ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
-                            struct ofproto_packet_in *pin)
+ofproto_dpif_send_async_msg(struct ofproto_dpif *ofproto,
+                            struct ofproto_async_msg *am)
 {
-    if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
+    if (!guarded_list_push_back(&ofproto->ams, &am->list_node, 1024)) {
         COVERAGE_INC(packet_in_overflow);
-        free(CONST_CAST(void *, pin->up.packet));
-        free(pin);
+        ofproto_async_msg_free(am);
     }
 
     /* Wakes up main thread for packet-in I/O. */
-    seq_change(ofproto->pins_seq);
+    seq_change(ofproto->ams_seq);
 }
 
 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
@@ -1236,7 +1235,7 @@ check_masked_set_action(struct dpif_backer *backer)
     return !error;
 }
 
-#define CHECK_FEATURE__(NAME, FIELD)                                        \
+#define CHECK_FEATURE__(NAME, SUPPORT, FIELD, VALUE)                        \
 static bool                                                                 \
 check_##NAME(struct dpif_backer *backer)                                    \
 {                                                                           \
@@ -1247,12 +1246,12 @@ check_##NAME(struct dpif_backer *backer)                                    \
     struct odp_flow_key_parms odp_parms = {                                 \
         .flow = &flow,                                                      \
         .support = {                                                        \
-            .NAME = true,                                                   \
+            .SUPPORT = true,                                                \
         },                                                                  \
     };                                                                      \
                                                                             \
     memset(&flow, 0, sizeof flow);                                          \
-    flow.FIELD = 1;                                                         \
+    flow.FIELD = VALUE;                                                     \
                                                                             \
     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);                         \
     odp_flow_key_from_flow(&odp_parms, &key);                               \
@@ -1267,12 +1266,13 @@ check_##NAME(struct dpif_backer *backer)                                    \
                                                                             \
     return enable;                                                          \
 }
-#define CHECK_FEATURE(FIELD) CHECK_FEATURE__(FIELD, FIELD)
+#define CHECK_FEATURE(FIELD) CHECK_FEATURE__(FIELD, FIELD, FIELD, 1)
 
 CHECK_FEATURE(ct_state)
 CHECK_FEATURE(ct_zone)
 CHECK_FEATURE(ct_mark)
-CHECK_FEATURE__(ct_label, ct_label.u64.lo)
+CHECK_FEATURE__(ct_label, ct_label, ct_label.u64.lo, 1)
+CHECK_FEATURE__(ct_state_nat, ct_state, ct_state, CS_TRACKED|CS_SRC_NAT)
 
 #undef CHECK_FEATURE
 #undef CHECK_FEATURE__
@@ -1293,6 +1293,8 @@ check_support(struct dpif_backer *backer)
     backer->support.odp.ct_zone = check_ct_zone(backer);
     backer->support.odp.ct_mark = check_ct_mark(backer);
     backer->support.odp.ct_label = check_ct_label(backer);
+
+    backer->support.odp.ct_state_nat = check_ct_state_nat(backer);
 }
 
 static int
@@ -1326,7 +1328,7 @@ construct(struct ofproto *ofproto_)
     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
     ovs_mutex_init(&ofproto->vsp_mutex);
 
-    guarded_list_init(&ofproto->pins);
+    guarded_list_init(&ofproto->ams);
 
     hmap_init(&ofproto->vlandev_map);
     hmap_init(&ofproto->realdev_vid_map);
@@ -1336,8 +1338,8 @@ construct(struct ofproto *ofproto_)
     sset_init(&ofproto->port_poll_set);
     ofproto->port_poll_errno = 0;
     ofproto->change_seq = 0;
-    ofproto->pins_seq = seq_create();
-    ofproto->pins_seqno = seq_read(ofproto->pins_seq);
+    ofproto->ams_seq = seq_create();
+    ofproto->ams_seqno = seq_read(ofproto->ams_seq);
 
 
     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
@@ -1403,8 +1405,7 @@ add_internal_flows(struct ofproto_dpif *ofproto)
     controller = ofpact_put_CONTROLLER(&ofpacts);
     controller->max_len = UINT16_MAX;
     controller->controller_id = 0;
-    controller->reason = OFPR_NO_MATCH;
-    ofpact_pad(&ofpacts);
+    controller->reason = OFPR_IMPLICIT_MISS;
 
     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
                                    &ofproto->miss_rule);
@@ -1442,10 +1443,10 @@ static void
 destruct(struct ofproto *ofproto_)
 {
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
-    struct ofproto_packet_in *pin;
+    struct ofproto_async_msg *am;
     struct rule_dpif *rule;
     struct oftable *table;
-    struct ovs_list pins;
+    struct ovs_list ams;
 
     ofproto->backer->need_revalidate = REV_RECONFIGURE;
     xlate_txn_start();
@@ -1465,12 +1466,11 @@ destruct(struct ofproto *ofproto_)
     }
     ofproto_group_delete_all(&ofproto->up);
 
-    guarded_list_pop_all(&ofproto->pins, &pins);
-    LIST_FOR_EACH_POP (pin, list_node, &pins) {
-        free(CONST_CAST(void *, pin->up.packet));
-        free(pin);
+    guarded_list_pop_all(&ofproto->ams, &ams);
+    LIST_FOR_EACH_POP (am, list_node, &ams) {
+        ofproto_async_msg_free(am);
     }
-    guarded_list_destroy(&ofproto->pins);
+    guarded_list_destroy(&ofproto->ams);
 
     recirc_free_ofproto(ofproto, ofproto->up.name);
 
@@ -1493,7 +1493,7 @@ destruct(struct ofproto *ofproto_)
     ovs_mutex_destroy(&ofproto->stats_mutex);
     ovs_mutex_destroy(&ofproto->vsp_mutex);
 
-    seq_destroy(ofproto->pins_seq);
+    seq_destroy(ofproto->ams_seq);
 
     close_dpif_backer(ofproto->backer);
 }
@@ -1512,23 +1512,22 @@ run(struct ofproto *ofproto_)
         mcast_snooping_mdb_flush(ofproto->ms);
     }
 
-    /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
+    /* Always updates the ofproto->ams_seqno to avoid frequent wakeup during
      * flow restore.  Even though nothing is processed during flow restore,
-     * all queued 'pins' will be handled immediately when flow restore
+     * all queued 'ams' will be handled immediately when flow restore
      * completes. */
-    ofproto->pins_seqno = seq_read(ofproto->pins_seq);
+    ofproto->ams_seqno = seq_read(ofproto->ams_seq);
 
     /* Do not perform any periodic activity required by 'ofproto' while
      * waiting for flow restore to complete. */
     if (!ofproto_get_flow_restore_wait()) {
-        struct ofproto_packet_in *pin;
-        struct ovs_list pins;
+        struct ofproto_async_msg *am;
+        struct ovs_list ams;
 
-        guarded_list_pop_all(&ofproto->pins, &pins);
-        LIST_FOR_EACH_POP (pin, list_node, &pins) {
-            connmgr_send_packet_in(ofproto->up.connmgr, pin);
-            free(CONST_CAST(void *, pin->up.packet));
-            free(pin);
+        guarded_list_pop_all(&ofproto->ams, &ams);
+        LIST_FOR_EACH_POP (am, list_node, &ams) {
+            connmgr_send_async_msg(ofproto->up.connmgr, am);
+            ofproto_async_msg_free(am);
         }
     }
 
@@ -1635,13 +1634,11 @@ wait(struct ofproto *ofproto_)
     mcast_snooping_wait(ofproto->ms);
     stp_wait(ofproto);
     if (ofproto->backer->need_revalidate) {
-        /* Shouldn't happen, but if it does just go around again. */
-        VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
         poll_immediate_wake();
     }
 
     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
-    seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
+    seq_wait(ofproto->ams_seq, ofproto->ams_seqno);
 }
 
 static void
@@ -3730,7 +3727,10 @@ ofproto_dpif_execute_actions__(struct ofproto_dpif *ofproto,
     xin.resubmit_stats = &stats;
     xin.recurse = recurse;
     xin.resubmits = resubmits;
-    xlate_actions(&xin, &xout);
+    if (xlate_actions(&xin, &xout) != XLATE_OK) {
+        error = EINVAL;
+        goto out;
+    }
 
     execute.actions = odp_actions.data;
     execute.actions_len = odp_actions.size;
@@ -3749,7 +3749,7 @@ ofproto_dpif_execute_actions__(struct ofproto_dpif *ofproto,
     execute.packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
 
     error = dpif_execute(ofproto->backer->dpif, &execute);
-
+out:
     xlate_out_uninit(&xout);
     ofpbuf_uninit(&odp_actions);
 
@@ -3899,13 +3899,13 @@ rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
     /* We always unwildcard nw_frag (for IP), so they
      * need not be unwildcarded here. */
     if (flow->nw_frag & FLOW_NW_FRAG_ANY
-        && ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
-        if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
+        && ofproto->up.frag_handling != OFPUTIL_FRAG_NX_MATCH) {
+        if (ofproto->up.frag_handling == OFPUTIL_FRAG_NORMAL) {
             /* We must pretend that transport ports are unavailable. */
             flow->tp_src = htons(0);
             flow->tp_dst = htons(0);
         } else {
-            /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
+            /* Must be OFPUTIL_FRAG_DROP (we don't have OFPUTIL_FRAG_REASM).
              * Use the drop_frags_rule (which cannot disappear). */
             rule = ofproto->drop_frags_rule;
             if (stats) {
@@ -4012,39 +4012,93 @@ rule_dealloc(struct rule *rule_)
 }
 
 static enum ofperr
-rule_check(struct rule *rule)
+check_mask(struct ofproto_dpif *ofproto, const struct miniflow *flow)
 {
+    const struct odp_support *support;
     uint16_t ct_state, ct_zone;
-    const ovs_u128 *labelp;
-    ovs_u128 ct_label = { { 0, 0 } };
+    ovs_u128 ct_label;
     uint32_t ct_mark;
 
-    ct_state = MINIFLOW_GET_U16(rule->cr.match.flow, ct_state);
-    ct_zone = MINIFLOW_GET_U16(rule->cr.match.flow, ct_zone);
-    ct_mark = MINIFLOW_GET_U32(rule->cr.match.flow, ct_mark);
-    labelp = MINIFLOW_GET_U128_PTR(rule->cr.match.flow, ct_label);
-    if (labelp) {
-        ct_label = *labelp;
+    support = &ofproto_dpif_get_support(ofproto)->odp;
+    ct_state = MINIFLOW_GET_U16(flow, ct_state);
+    if (support->ct_state && support->ct_zone && support->ct_mark
+        && support->ct_label && support->ct_state_nat) {
+        return ct_state & CS_UNSUPPORTED_MASK ? OFPERR_OFPBMC_BAD_MASK : 0;
+    }
+
+    ct_zone = MINIFLOW_GET_U16(flow, ct_zone);
+    ct_mark = MINIFLOW_GET_U32(flow, ct_mark);
+    ct_label = MINIFLOW_GET_U128(flow, ct_label);
+
+    if ((ct_state && !support->ct_state)
+        || (ct_state & CS_UNSUPPORTED_MASK)
+        || ((ct_state & (CS_SRC_NAT | CS_DST_NAT)) && !support->ct_state_nat)
+        || (ct_zone && !support->ct_zone)
+        || (ct_mark && !support->ct_mark)
+        || (!ovs_u128_is_zero(&ct_label) && !support->ct_label)) {
+        return OFPERR_OFPBMC_BAD_MASK;
     }
 
-    if (ct_state || ct_zone || ct_mark
-        || !ovs_u128_is_zero(&ct_label)) {
-        struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->ofproto);
-        const struct odp_support *support = &ofproto_dpif_get_support(ofproto)->odp;
+    return 0;
+}
+
+static enum ofperr
+check_actions(const struct ofproto_dpif *ofproto,
+              const struct rule_actions *const actions)
+{
+    const struct ofpact *ofpact;
+
+    OFPACT_FOR_EACH (ofpact, actions->ofpacts, actions->ofpacts_len) {
+        const struct odp_support *support;
+        const struct ofpact_conntrack *ct;
+        const struct ofpact *a;
+
+        if (ofpact->type != OFPACT_CT) {
+            continue;
+        }
+
+        ct = CONTAINER_OF(ofpact, struct ofpact_conntrack, ofpact);
+        support = &ofproto_dpif_get_support(ofproto)->odp;
 
-        if ((ct_state && !support->ct_state)
-            || (ct_zone && !support->ct_zone)
-            || (ct_mark && !support->ct_mark)
-            || (!ovs_u128_is_zero(&ct_label) && !support->ct_label)) {
-            return OFPERR_OFPBMC_BAD_FIELD;
+        if (!support->ct_state) {
+            return OFPERR_OFPBAC_BAD_TYPE;
         }
-        if (ct_state & CS_UNSUPPORTED_MASK) {
-            return OFPERR_OFPBMC_BAD_MASK;
+        if ((ct->zone_imm || ct->zone_src.field) && !support->ct_zone) {
+            return OFPERR_OFPBAC_BAD_ARGUMENT;
+        }
+
+        OFPACT_FOR_EACH(a, ct->actions, ofpact_ct_get_action_len(ct)) {
+            const struct mf_field *dst = ofpact_get_mf_dst(a);
+
+            if (a->type == OFPACT_NAT && !support->ct_state_nat) {
+                /* The backer doesn't seem to support the NAT bits in
+                 * 'ct_state': assume that it doesn't support the NAT
+                 * action. */
+                return OFPERR_OFPBAC_BAD_TYPE;
+            }
+            if (dst && ((dst->id == MFF_CT_MARK && !support->ct_mark)
+                        || (dst->id == MFF_CT_LABEL && !support->ct_label))) {
+                return OFPERR_OFPBAC_BAD_SET_ARGUMENT;
+            }
         }
     }
+
     return 0;
 }
 
+static enum ofperr
+rule_check(struct rule *rule)
+{
+    struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->ofproto);
+    enum ofperr err;
+
+    err = check_mask(ofproto, &rule->cr.match.mask->masks);
+    if (err) {
+        return err;
+    }
+    return check_actions(ofproto, rule->actions);
+}
+
 static enum ofperr
 rule_construct(struct rule *rule_)
     OVS_NO_THREAD_SAFETY_ANALYSIS
@@ -4343,10 +4397,10 @@ get_datapath_version(const struct ofproto *ofproto_)
 
 static bool
 set_frag_handling(struct ofproto *ofproto_,
-                  enum ofp_config_flags frag_handling)
+                  enum ofputil_frag_handling frag_handling)
 {
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
-    if (frag_handling != OFPC_FRAG_REASM) {
+    if (frag_handling != OFPUTIL_FRAG_REASM) {
         ofproto->backer->need_revalidate = REV_RECONFIGURE;
         return true;
     } else {
@@ -4540,7 +4594,7 @@ ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
                                    name, sizeof name);
             ds_put_format(&ds, "%5s  %4d  ", name, grp->vlan);
-            print_ipv6_mapped(&ds, &grp->addr);
+            ipv6_format_mapped(&grp->addr, &ds);
             ds_put_format(&ds, "         %3d\n",
                           mcast_bundle_age(ofproto->ms, b));
         }
@@ -4993,6 +5047,7 @@ ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
               struct ds *ds)
 {
     struct trace_ctx trace;
+    enum xlate_error error;
 
     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
     ds_put_cstr(ds, "Flow: ");
@@ -5012,8 +5067,7 @@ ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
     trace.xin.resubmit_hook = trace_resubmit;
     trace.xin.report_hook = trace_report_valist;
 
-    xlate_actions(&trace.xin, &trace.xout);
-
+    error = xlate_actions(&trace.xin, &trace.xout);
     ds_put_char(ds, '\n');
     trace_format_flow(ds, 0, "Final flow", &trace);
     trace_format_megaflow(ds, 0, "Megaflow", &trace);
@@ -5021,7 +5075,10 @@ ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
     ds_put_cstr(ds, "Datapath actions: ");
     format_odp_actions(ds, trace.odp_actions.data, trace.odp_actions.size);
 
-    if (trace.xout.slow) {
+    if (error != XLATE_OK) {
+        ds_put_format(ds, "\nTranslation failed (%s), packet is dropped.\n",
+                      xlate_strerror(error));
+    } else if (trace.xout.slow) {
         enum slow_path_reason slow;
 
         ds_put_cstr(ds, "\nThis flow is handled by the userspace "
@@ -5267,6 +5324,8 @@ disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
         ofproto_use_tnl_push_pop = true;
         unixctl_command_reply(conn, "Tunnel push-pop on");
         ofproto_revalidate_all_backers();
+    } else {
+        unixctl_command_reply_error(conn, "Invalid argument");
     }
 }
 
@@ -5597,22 +5656,17 @@ ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
     struct rule_dpif *rule;
     int error;
 
-    ofm.fm.match = *match;
-    ofm.fm.priority = priority;
-    ofm.fm.new_cookie = htonll(0);
-    ofm.fm.cookie = htonll(0);
-    ofm.fm.cookie_mask = htonll(0);
-    ofm.fm.modify_cookie = false;
-    ofm.fm.table_id = TBL_INTERNAL;
-    ofm.fm.command = OFPFC_ADD;
-    ofm.fm.idle_timeout = idle_timeout;
-    ofm.fm.hard_timeout = 0;
-    ofm.fm.importance = 0;
-    ofm.fm.buffer_id = 0;
-    ofm.fm.out_port = 0;
-    ofm.fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
-    ofm.fm.ofpacts = ofpacts->data;
-    ofm.fm.ofpacts_len = ofpacts->size;
+    ofm.fm = (struct ofputil_flow_mod) {
+        .match = *match,
+        .priority = priority,
+        .table_id = TBL_INTERNAL,
+        .command = OFPFC_ADD,
+        .idle_timeout = idle_timeout,
+        .flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY,
+        .ofpacts = ofpacts->data,
+        .ofpacts_len = ofpacts->size,
+        .delete_reason = OVS_OFPRR_NONE,
+    };
 
     error = ofproto_flow_mod(&ofproto->up, &ofm);
     if (error) {
@@ -5641,15 +5695,13 @@ ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
     struct ofproto_flow_mod ofm;
     int error;
 
-    ofm.fm.match = *match;
-    ofm.fm.priority = priority;
-    ofm.fm.new_cookie = htonll(0);
-    ofm.fm.cookie = htonll(0);
-    ofm.fm.cookie_mask = htonll(0);
-    ofm.fm.modify_cookie = false;
-    ofm.fm.table_id = TBL_INTERNAL;
-    ofm.fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
-    ofm.fm.command = OFPFC_DELETE_STRICT;
+    ofm.fm = (struct ofputil_flow_mod) {
+        .match = *match,
+        .priority = priority,
+        .table_id = TBL_INTERNAL,
+        .flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY,
+        .command = OFPFC_DELETE_STRICT,
+    };
 
     error = ofproto_flow_mod(&ofproto->up, &ofm);
     if (error) {